Drug interactions are not actually a data problem. This is a contrarian take, especially when building a health AI that relies heavily on medical knowledge. On paper, it sounds simple: get a database of drugs, list interactions, flag them. Medical schools, in their compressed curriculum, might dedicate a week to the topic, making it seem like a finite set of facts to memorize. But as I've spent the better part of three months just modeling them for GoDavaii, I can tell you the real challenge isn't the data quantity, it's the contextual quality.
Beyond the 'Fact': Why Context Matters More Than Count
When we started GoDavaii, our vision was clear: India's Advanced Health AI. And 'advanced' meant going beyond just English. It meant understanding the nuances of how a family in Chennai discusses a medicine versus one in Kolkata, in their own mother tongue. This immediately threw a wrench into the 'simple data problem' idea. How do you flag an interaction when one medicine is known by three different brand names in Tamil, and the other has a common 'desi ilaaj' (home remedy) that also interacts, but isn't in any Western database?
My initial approach, like many, was to aggregate. Pull in every reputable drug database. Standardize. Match. But it quickly became evident that this was like trying to understand a conversation by only reading keywords. Take for instance, a patient taking a common allopathic drug for blood pressure, and simultaneously using an Ayurvedic formulation for general well-being. Mainstream drug interaction checkers, built primarily for Western markets, have no framework for the latter. Our AI-verified Desi Ilaaj feature isn't just a separate module; it's deeply interwoven with the interaction checker to bridge this gap.
Architecting for 22+ Languages and Cultural Context
The real deep dive began when we started building the AI Health Chat in 22+ Indian languages. It's one thing to translate drug names; it's another entirely to understand the intent behind a user's query about a medicine in Marathi, or to parse a symptom description in Gujarati. For example, tabiyat theek nahi in Tamil translates to 'not feeling well', but depending on context, it could imply anything from mild discomfort to severe illness. Our interaction checker needs to consider these linguistic and cultural variations not just in drug names, but in how symptoms are described, how home remedies are discussed, and even the common food interactions that might not be clinically documented in a pharma textbook.
We use state-of-the-art LLMs, like Gemini 2.5 Flash, fine-tuned specifically on Indian medical corpora to handle the linguistic complexity. But the core interaction logic is a custom knowledge graph. It's not just drug_A -> interacts_with -> drug_B. It's drug_A -> (metabolized_by: CYP3A4) -> (side_effect: Drowsiness), and desi_ilaaj_X -> (active_compound: Curcumin) -> (metabolized_by: CYP3A4). The graph then dynamically evaluates potential overlaps, risks, and even contraindications, presenting them in a way a family can understand, not just a doctor.
This is why, as a founder, I often describe GoDavaii as building for the next billion - the people coming online in their mother tongue, with health questions English AI simply cannot answer. We placed Top 14 Global at Startup Flight Vietnam 2025, and honestly, the sheer scale of the language challenge was something few global VCs truly grasped. It's a genuine moat, built on layers of linguistic and medical intelligence.
GoDavaii as a Thinking Tool, Not a Replacement
Building this isn't about replacing doctors. It's about empowering families. It's a family-side assistant before the appointment. When a new drug for advanced ovarian cancer makes headlines (like the recent NHS drug news), or discussions around GLP-1 receptor agonists become prevalent, the complexity of managing multiple medications only grows. Our tool helps surface those crucial questions, catches what a 7-minute appointment might miss, and gives families a more informed position to discuss with their physician.
This sprint has been a deep dive into refining these models. The journey of building GoDavaii is less about grand gestures and more about getting these complex details right, one language, one interaction, one cultural nuance at a time. It's hard, slow work, but it's the only way to build truly advanced health AI.
--Pururva Agarwal, Founder, GoDavaii. Learn more at https://www.godavaii.com
Top comments (0)